AI will have a disruptive impact on procurement technology. Application leaders in procurement need to ensure access to the right skills and data to be able to leverage these new capabilities. This research provides recommendations on how to adopt and get value from AI in procurement.
Application leaders responsible for modernizing procurement applications should:
In 2020, modest adoption of procurement and sourcing technology will limit smart machine technology's penetration to 10% of the total available market.
By 2020, automation and smart machines will reduce employee requirements in business shared-service centers by 65%.
Basic machine learning technology is today already used by some procurement vendors in areas such as spend analytics and contract analytics — see use cases in the Impact and Recommendations section. However, procurement technology vendors are now starting to create cognitive expert advisors (CEAs) and virtual personal assistants (VPAs) to further automate procurement processes, and increase the efficiency and effectiveness of procurement organizations. Indeed, vendors are now leveraging more advanced machine learning and machine-learning-based technologies such as natural-language processing (NLP) and natural-language generation (NLG). For example, procurement VPAs can improve the user experience of transactional procurement tools and increase spend under management by guiding end users to the right purchasing tool. In the strategic sourcing space, cognitive procurement advisors (CPAs) can be used to provide summaries, recommendations and advice in areas such as predictive analytics, sourcing award scenarios, supplier and proposal assessments, risk management and supplier performance management. This is a result of the unprecedented advances in smart machines and artificial intelligence (AI) we have seen in recent years, thanks to developments in hardware, algorithms and data availability.
AI, however, is a vague concept and the term is used quite liberally by vendors in various marketing activities. In most cases, AI refers to different sorts of machine learning capabilities with different degrees of sophistication. The leading edge of machine learning today is deep learning (also called deep neural networks [DNNs]). With deep learning, a computer model can be fed lots of complex data, such as images, speech and text. For example, deep-learning algorithms can analyze retina scans to independently "figure out" which patterns indicate healthy or diseased retinas (and to indicate the specific disease). The "figuring out" process relies on brute force and high-performance computing, and can, to some extent, render obsolete the tedious handcrafting of features and data preparation.
Source: Gartner (March 2017)
Before most organizations will get any significant value from AI in procurement (apart from some very narrow scopes like spend analytics, contract analytics or risk management; more details on this can be found in the next section), there is a significant issue to address: adoption. Most organizations only manage part of their spend with procurement technology. In the "Magic Quadrant for Strategic Sourcing Application Suites" reference survey from 2016, almost 50% of client references responded that they run fewer than 100 sourcing events per year and almost 60% run fewer than 10 reverse auctions per year. In the transactional procurement space, many organizations also struggle with automating spend management. This is due to poor implementations and change management, the use of inflexible legacy systems with poor UIs, or having large spend categories for which the tools in place are not well-suited.
The first step to drive automation of spend management is to make sure that existing tools, if deemed suitable, are used by the organization. Responsibility for the consistent use and validation of spend analytics data should be allocated to category managers (or the equivalent). Also, the use of e-sourcing should be mandatory, even for renegotiations and negotiations with a single supplier, to ensure that data is captured (for more details on e-sourcing, see "Adopt These E-Sourcing Best Practices to Drive Savings"). All contracts should be, at a minimum, stored in a contract repository. Procure-to-pay (P2P) tools should be properly configured and populated by relevant suppliers. In addition, other purchase order (PO) source systems should be identified, and integrated if necessary, to make sure that end users can find what they need and that automatic invoice matching is enabled as far as possible. Proper use of existing tools provides the platform for applying AI, as well as providing the data on which to train the AI.
The second step is to address spend categories that are not suitable to be managed through existing solutions. Automating 100% of spend management is not realistic nor necessary for most organizations, but it should be a strategic choice to not manage parts of your spend.
In the e-sourcing space, many general-purpose tools struggle to deal with complex spend categories. This may be due to the large numbers of line items to be negotiated, complex evaluation scenarios, or the need to evaluate alternative bids and the implications they may have. Classic examples of this are logistics, packaging and, in some cases, manufacturing materials. Organizations with significant spend in these or similar areas should investigate advanced sourcing solutions such as Trade Extensions' TESS, Jaggaer's Advanced Sourcing Optimizer, Keelvar and BravoAdvantage SourcingPlus.
In the transactional procurement area, general-purpose P2P solutions are primarily designed for PO and catalog-based spend — that is, what you order can be prenegotiated and a price can be set on the PO that the invoice can later be matched against. However, in many spend categories, you might not know what the cost will be until, for example, the service has been performed. It could also be the case that a PO might not add any value (for example, in the case of a lease or subscription). Buying organizations need to analyze their spend from an ordering and invoicing characteristics point of view, and deploy a workstream-specific P2P strategy to make sure that the organization has the appropriate tools (for more details, see "Procurement Leaders Need to Deploy Work-Stream-Specific P2P Solutions to Maximize ROI").
Better adoption of these procurement tools will enable maturing AI capabilities to improve processes, provide actionable advice and increase automation. It will also improve the access to data that is critical for AI — see more on this in the last section.
Over the last decade, many procurement organizations have shifted from a traditional, negotiation-focused approach to a category management approach, with a greater focus on using technology, and on activities such as demand management, change management and supplier innovation. This shift has required different skills and has resulted in many organizations needing to retrain existing staff, as well as recruit new talent.
The emergence of machine learning and AI is introducing the need for analytical skills and an understanding of data science and technology. Organizations that don't already have dedicated procurement analysts or procurement technology experts will need to create this role to make sure they understand how and when to leverage AI technology in procurement. If you are investing in procurement technology, you need to have someone that understands the impacts and requirements of AI. However, smaller organizations with less than $500 million in spend might not be able to justify a full-time procurement analyst.
Progressive procurement organizations need to start addressing the role of AI in procurement now. Since data scientists are in short supply and unlikely to work as procurement analysts (and hiring one would probably be over kill), they should be trained to become "citizen (procurement) data scientists." Time and budget should be set aside to provide at least basic training for your procurement analysts; there are both self-learning courses (for example, on YouTube) and data science courses available.
Once the resources, basic skills and understanding are in place, procurement organizations should experiment and run pilots in areas where there are existing use cases. Three areas to explore are:
The different use cases require different levels of understanding of AI. In the spend analytics case, an understanding of what type of data is useful for the solution and can provide additional value is enough. Then the spend analytics vendor applies the AI capabilities. In the contract analytics and risk management areas, a deeper understanding is necessary, because it is important to have the quantitative skills to figure out what to analyze or measure, and how to score it. For some use cases, no AI skills whatsoever are necessary. For example, in the invoice management space, AI is also used across multiple vendors. While e-invoicing is on the rise, many organization still have suppliers that are not ready to send "true e-invoices." However, they are likely to at least be able to send invoices in PDF format. Some P2P and accounts payables invoice automation (APIA) vendors, such as Coupa, Proactis, Lexmark and Basware, today have machine learning technologies to help parse, interpret and extract data from PDFs or scanned documents to enable automated matching or approval workflows. These solutions learn over time and gradually become more proficient at correctly extracting the relevant data. This reduces the need for manual data transfer and shortening invoice cycle times. Here, AI capabilities are used in more of a "black box" mode, leaving little room for experimentation.
Other areas where AI is having an impact include the contingent workforce and freelance management. Indeed, vendors such as Nétive VMS and DCR Workforce are using machine learning to improve CV matching and candidate selection. There are also opportunities in areas such as supplier information management and master data management to match and check data points.
Even though great advances have been made in AI, the tools still need data to work with — in most cases, a lot of data. While a child can learn to recognize an object after being told what it is only a few times, a deep-learning system may need thousands of examples and to "look" at those examples hundreds of thousands or millions of times before getting it right. This means that the more advanced AI applications in particular, such as CEAs or VPAs, will need to be trained on very large and reliable datasets in order to provide accurate and relevant recommendations, and the right assistance. As a procurement organization, you need to understand what data you need and what data you have access to. This data can be broadly categorized into four main groups: internally generated data, supplier data, publicly available data and subscription data (see Figure 2).
Source: Gartner (March 2017)
Organizations with low adoption of integrated procurement solutions and a fragmented solutions landscape can consider using robotic process automation (RPA) tools as an interim option to facilitate better procurement practices, and access to supplier and internal data. The use cases for RPA tools in procurement include:
RPA tools are not "intelligent." RPA needs to be paired with tools to structure unstructured data. It can be considered for situations in which organizations have found that other integration or automation options are perhaps too expensive or too time-consuming (taking months or years). It should be considered both in light of other technical options and the need for some process change management skills. Use cases for RPA include instances in which an organization wants to work with structured data to:
RPA tools mimic the same "manual" path taken through applications by a human using a combination of UI interactions, APIs, mainframe and client server metadata tags. A process is not synonymous with a robot, but rather a "pool" of robots are capable of executing a task until it is fulfilled and can then execute a second task.
What data you need is of course dependent on what you want to achieve. For instance, if you want to create a cognitive procurement advisor to help evaluate supplier performance, you might want quality data from a quality management system or delivery performance data from your ERP system, but how about less tangible information, such as supplier innovation and customer service? Is this type of information captured anywhere and can it be extracted? To maximize the availability of data, processes need to be digitized as far as possible to make sure that as much data as possible is available in electronic formats. After all, it's hard for even the smartest AI to find that paper contract in your file cabinet or the notes from the last supplier meeting in your paper notebook.
For buying organization looking to explore AI technologies, understanding what data is needed and ensuring access to it will be critical.
Source: Gartner Research Note G00325362, Magnus Bergfors, 27 March 2017